|
From: | BALATON Zoltan |
Subject: | Re: [PATCH 3/3] sam460ex: Use type cast macro instead of simple cast |
Date: | Thu, 7 Jan 2021 20:37:52 +0100 (CET) |
On Thu, 7 Jan 2021, Greg Kurz wrote:
On Thu, 7 Jan 2021 10:45:26 +0100 BALATON Zoltan <balaton@eik.bme.hu> wrote:On Thu, 7 Jan 2021, Greg Kurz wrote:On Wed, 6 Jan 2021 16:24:01 +0100 BALATON Zoltan via <qemu-ppc@nongnu.org> wrote:Use the PCI_BUS type cast macro to convert result of qdev_get_child_bus(). Also remove the check for NULL afterwards which should not be needed because sysbus_create_simple() uses error_abortIt seems to me that sysbus_create_simple() doesn't return NULL because it ends up calling object_new_with_type(). This allocates the object with either g_malloc() or qemu_memalign(), both of which abort on failure.and PCI_BUS macro also checks its argument by default so thisAFAICT, PCI_BUS() and all other instance type checking macros are happy with a NULL argument. They simply return NULL in this case.This wasn't my experience when I've got an error in code and got a NULL pointer here (on pegasos2 board but same situation). At least with qom-debug enabled (which I think is on by default unless explicitly disabled in configure) this will abort if the object is not the right type.You're right that qom-cast-debug is enabled by default and that it causes object_dynamic_cast_assert() to abort on type mismatch, but definitely not with a NULL value, as mentioned in this very old commit:
Indeed, PCI_BUS(NULL) does not abort just returns NULL. I think I remembered wrong and had dev==NULL so qdev_get_child_bus() was aborting.
commit b7f43fe46029d8fd0594cd599fa2599dcce0f553 Author: Paolo Bonzini <pbonzini@redhat.com> Date: Fri Nov 23 16:56:17 2012 +0100 qom: dynamic_cast of NULL is always NULL Trying to cast a NULL value will cause a crash. Returning NULL is also sensible, and it is also what the type-unsafe DO_UPCAST macro does. Reported-by: Markus Armbruster <armbru@redhat.com> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Anthony Liguori <aliguori@us.ibm.com> Maybe this should be documented in the function header in "qom/object.h".shouldn't fail here. Signed-off-by: BALATON Zoltan <balaton@eik.bme.hu> --- hw/ppc/sam460ex.c | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-) diff --git a/hw/ppc/sam460ex.c b/hw/ppc/sam460ex.c index 14e6583eb0..cc67e9c39b 100644 --- a/hw/ppc/sam460ex.c +++ b/hw/ppc/sam460ex.c @@ -384,11 +384,8 @@ static void sam460ex_init(MachineState *machine) ppc460ex_pcie_init(env); /* All PCI irqs are connected to the same UIC pin (cf. UBoot source) */ dev = sysbus_create_simple("ppc440-pcix-host", 0xc0ec00000, uic[1][0]); - pci_bus = (PCIBus *)qdev_get_child_bus(dev, "pci.0"); - if (!pci_bus) { - error_report("couldn't create PCI controller!"); - exit(1); - } + pci_bus = PCI_BUS(qdev_get_child_bus(dev, "pci.0")); +But PCI_BUS() is being passed qdev_get_child_bus(dev, "pci.0"), not dev... so the real question here is whether this can return NULL or not. And if this happens, is this a (1) user or (2) programming error ? If (1) then the "if (!pci_bus) { }" should be kept. If (2) then it should be converted to an assert().I think it can only fail if the ppc440-pcix-host type is changed to not have a pci.0 child any more which is a programming error that's very unlikely to happen but if needed an assert could be added but I don't think that's really necessary. The error_report was definitely not needed as it's not a user error in any case.I was also thinking about a programming error. Whether to add an assert() or not is up to you, you're the maintainer for this code :)
In that case I think I keep it simple and don't add an assert because I think this error is highly unlikely (we create a pci host object that should have a pci bus child) and it would crash anyway shortly when trying to add devices so an additional assert here does not seem to help much catching a bug.
Regards, BALATON Zoltan
[Prev in Thread] | Current Thread | [Next in Thread] |